Scholarlink

<p>Reliable arrival time prediction is essential for the efficient operation of passenger rail systems, particularly in regional networks where journeys frequently involve transfers between multiple train services. Delays in rail operations can propagate across the network and significantly affect passengers’ ability to catch connecting services and arrive at their final destination on time. While many existing studies focus on predicting delays for individual train services, predicting journey-level arrival times across connected legs remains challenging because of transfer interactions at interchange stations and heterogeneous operational dynamics between different services. This paper proposes a hierarchical predictive modeling framework for estimating arrival delays in passenger rail journeys with intra-modal transfers. The framework combines machine learning models developed for individual legs of a journey and integrates their outputs through a corridor-level meta-model to estimate the final arrival delay at the destination. Leg-level predictions are generated using Extreme Gradient Boosting (XGBoost), which captures nonlinear relationships between operational, temporal, and environmental variables. These predictions are then aggregated in a hierarchical structure that accounts for the interaction between upstream delays and transfer feasibility at interchange stations. Transfer feasibility is derived from the predicted arrival time of the feeder service, the scheduled departure time of the connecting service, and the minimum transfer margin required at the interchange. The framework is evaluated using real-world operational data from a regional rail corridor consisting of two connected services linked by an interchange station. The results show that the aggregated approach improves journey-level delay prediction relative to leg-specific prediction alone. Moreover, explicitly modeling transfer feasibility allows the framework to capture the operational consequences of maintained and missed connections. Overall, the findings demonstrate that hierarchical aggregation of leg-level predictive models can improve journey-level reliability estimation and provide useful decision support for regional passenger rail operations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A hierarchical machine learning framework for aggregated multi-leg arrival time prediction and transfer-feasibility integration in regional rail corridors

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月16日

查看全文
  • Elham Ahmadi,
  • André Ludwig,
  • Henrik Leopold

摘要

Reliable arrival time prediction is essential for the efficient operation of passenger rail systems, particularly in regional networks where journeys frequently involve transfers between multiple train services. Delays in rail operations can propagate across the network and significantly affect passengers’ ability to catch connecting services and arrive at their final destination on time. While many existing studies focus on predicting delays for individual train services, predicting journey-level arrival times across connected legs remains challenging because of transfer interactions at interchange stations and heterogeneous operational dynamics between different services. This paper proposes a hierarchical predictive modeling framework for estimating arrival delays in passenger rail journeys with intra-modal transfers. The framework combines machine learning models developed for individual legs of a journey and integrates their outputs through a corridor-level meta-model to estimate the final arrival delay at the destination. Leg-level predictions are generated using Extreme Gradient Boosting (XGBoost), which captures nonlinear relationships between operational, temporal, and environmental variables. These predictions are then aggregated in a hierarchical structure that accounts for the interaction between upstream delays and transfer feasibility at interchange stations. Transfer feasibility is derived from the predicted arrival time of the feeder service, the scheduled departure time of the connecting service, and the minimum transfer margin required at the interchange. The framework is evaluated using real-world operational data from a regional rail corridor consisting of two connected services linked by an interchange station. The results show that the aggregated approach improves journey-level delay prediction relative to leg-specific prediction alone. Moreover, explicitly modeling transfer feasibility allows the framework to capture the operational consequences of maintained and missed connections. Overall, the findings demonstrate that hierarchical aggregation of leg-level predictive models can improve journey-level reliability estimation and provide useful decision support for regional passenger rail operations.